Strategic Learning In Recommendation Systems

Effective personalization can help firms reduce their customers’ search costs and enhance customer loyalty. The personalization process consists of two important activities: learning and matching. Learning involves collecting data from a customer’s interactions with the firm and then making inferences from the data about the customer’s profile. Matching requires identifying which products to recommend or links to provide for making a sale. Prior research has typically looked at each activity in isolation. For instance, recent research has studied how a user’s profile can be inferred quickly by offering items (links) that help discriminate user classes. Research on matching has typically assumed that all the recommendations in an interaction are made to generate immediate sales. We examine the problem of identifying items to offer such that both learning and matching are taken into consideration, thereby enabling the firm to achieve higher payoffs in the long run.